is the low-complexity mobile-relay-aided ffr-das capable ...sxn/paper/201604tvt.pdf · cooperation,...

2154 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 65, NO. 4, APRIL 2016

Is the Low-Complexity Mobile-Relay-AidedFFR-DAS Capable of Outperforming the

High-Complexity CoMP?Shaoshi Yang, Member, IEEE, Xinyi Xu, Dimitrios Alanis, Student Member, IEEE,

Soon Xin Ng, Senior Member, IEEE, and Lajos Hanzo, Fellow, IEEE

Abstract—The coordinated multipoint transmission/receptionaided collocated antenna system (CoMP-CAS) and the mobilerelay assisted fractional frequency reuse distributed antenna sys-tem (MR-FFR-DAS) constitute a pair of virtual-MIMO-basedtechnical options for achieving high spectral efficiency in inter-ference-limited cellular networks. In practice, both techniqueshave their respective pros and cons, which are studied in thispaper by evaluating the achievable cell-edge performance on theuplink of multicell systems. We show that, assuming the sameantenna configuration in both networks, the maximum availablecooperative spatial diversity (or the multiplexing gain) inherent inthe MR-FFR-DAS is lower than that in the CoMP-CAS. However,when the cell-edge mobile stations (MSs) have a low transmissionpower, the lower complexity MR-FFR-DAS relying on the simplesingle-cell processing may outperform the CoMP-CAS by usingthe proposed soft-combining-based probabilistic data association(SC-PDA) receiver, despite the fact that the latter scheme is morecomplex and incurs a higher cooperation overhead. Furthermore,the benefits of the SC-PDA receiver may be enhanced by prop-erly selecting the MRs’ positions. Additionally, we show thatthe performance of the cell-edge MSs roaming near the angulardirection halfway between two adjacent remote antennas (i.e.,the “worst-case direction”) of the MR-FFR-DAS may be moresignificantly improved than that of the cell-edge MSs of otherdirections by using multiuser power control, which also improvesthe fairness among cell-edge MSs. Our simulation results showthat, given a moderate MS transmit power, the proposed MR-FFR-DAS architecture employing the SC-PDA receiver is capableof achieving significantly better bit error rate (BER) and effectivethroughput across the entire cell-edge area, including even theworst-case direction and the cell-edge boundary, than the CoMP-CAS architecture.

Manuscript received July 30, 2014; revised December 18, 2014 andFebruary 19, 2015; accepted March 21, 2015. Date of publication March 25,2015; date of current version April 14, 2016. This work was supported inpart by Research Councils UK through the India-UK Advanced TechnologyCentre, by the European Research Council’s Advanced Fellow Grant, by theEuropean Union through the CONCERTO Project, and by the Chinese National863 Program Project under Grant 2015AA10302. The review of this paper wascoordinated by Dr. C. Yuen.

S. Yang, D. Alanis, S. X. Ng, and L. Hanzo are with the School of Electronicsand Computer Science, University of Southampton, Southampton SO17 1BJ,U.K. (e-mail: [email protected]; [email protected]; [email protected]; [email protected]).

X. Xu was with the School of Electronics and Computer Science, Univer-sity of Southampton, Southampton SO17 1BJ, U.K. She is now with ChinaAcademy of Electronics and Information Technology, Beijing 100041, China(e-mail: [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TVT.2015.2416333

Index Terms—Base station (BS) cooperation, coordinated mul-tipoint (CoMP), distributed antenna system (DAS), fractional fre-quency reuse, mobile relay (MR), multicell uplink.

I. INTRODUCTION

FUTURE mobile communication systems are expected toprovide higher data rates and more homogeneous quality

of service (QoS) across the entire network. To meet thesedemands, various technical options have been suggested, whichinclude but are not limited to using more spectrum (e.g.,millimeter wave), using more antennas [e.g., massive multiple-input–multiple-output (MIMO) systems], and introducing ded-icated relays and small cells to form heterogeneous networks[1]. Although these options are promising, most of them re-quire installing additional equipment and a radical change ofthe network architecture and entities, which may be costlyfor the operators. From a pragmatic perspective, in the shortand medium terms, it may be more promising to upgrade theexisting cellular architecture using an evolutionary strategy.Hence, in this paper, we aim for investigating the pros and consof a pair of representative cellular architectures, namely, thecoordinated multipoint transmission/reception [2]–[11] aidedcollocated antenna system (CoMP-CAS) and the fractionalfrequency reuse [12]–[14] assisted distributed antenna systemrelying on mobile relays (MR-FFR-DAS), in the context of themulticell uplink. Both of them have the potential of providinga significant gain without incurring dramatic changes of theexisting cellular systems; hence, they are of great interest toboth industry and academia.

Benefits and Challenges of CoMP-CAS: On one hand, theconventional cellular architecture that relies on a CAS at eachbase station (BS) is still widely used, where the mobile stations(MSs) roaming in the cell-edge area typically suffer from alow throughput and low power efficiency. This is because lowsignal-to-interference-plus-noise ratio (SINR) may be experi-enced by the cell-edge MSs owing to the combined effectsof intercell cochannel interference (CCI) and path loss. As aremedy, CoMP techniques [2]–[11] have been advocated forthe CAS-based cellular architecture in the 3GPP Long-TermEvolution Advanced (LTE-A) standard, where the intercell CCIcan be mitigated or even beneficially exploited by cooperationamong the different sectors or cells.

However, to enhance the uplink performance of the cell-edgeMSs, a CoMP-aided CAS requires these MSs to transmit at

0018-9545 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

YANG et al.: IS THE MOBILE-RELAY-AIDED FFR-DAS CAPABLE OF OUTPERFORMING THE COMP? 2155

a rather high power [6], [15], [16]. Essentially, CoMP on theuplink relies on the joint decoding philosophy, and it achieves acooperative spatial diversity gain with the aid of the collabora-tive adjacent BSs equipped with CASs, which are quite far fromthe cell-edge MSs. Additionally, the CoMP techniques typi-cally require exchanging a significant amount of data/channelinformation among the cooperative entities, which results in apotentially excessive overhead traffic on the backhaul. Further-more, practical impairments, such as the asynchronous natureof intercell CCI [17], the backhaul capacity limitation, thechannel estimation inaccuracy, and the low channel coherencetime of the network-wide system [8]–[11], [18], [19], may alsosignificantly degrade the practically achievable benefits of theCoMP techniques.

Benefits and Challenges of MR-FFR-DAS: On the otherhand, the FFR philosophy [13], [14], [20]–[22], which confinesthe geographic scattering of the intercell CCI at the cost of amoderately reduced degree of frequency reuse, has been alsosuggested for the LTE initiative [12]. Furthermore, the largepath loss experienced by the cell-edge MSs may be reducedby employing DASs, where the remote antennas (RAs) arepositioned closer to the cell-edge MSs and connected to theBS using optical fiber [23]–[25]. Thus, the cell-edge MSs cantransmit at a relatively low power. Additionally, it is possibleto invoke the existing MSs as MRs [16], [26], which mayprovide additional benefits (reduced path loss and increasedspatial diversity gain) for cell-edge MSs. Hence, for a cellularsystem that cannot afford the more complex BS-cooperation-aided CoMP, we suggest that it might be a practically attractivesolution to amalgamate the benefits of FFR, DAS, and MRs ineach single cell for the sake of increasing the SINR at the RAs(on the uplink) or at the cell-edge MSs (on the downlink).

The MR-FFR-DAS also faces particular challenges imposedby the system architecture. In the MR-FFR-DAS, each cell-edge MS is served mainly by a nearby RA, whereas the otherRAs cannot provide the same level of support, since they are faraway from the cell-edge MS considered. As a result, althoughthe cell-edge MSs roaming close to the RAs do indeed benefitfrom a high SINR, there exist undesirable scenarios, wherethese cell-edge MSs suffer from increased intracell CCI. Morespecifically, when a cell-edge MS roaming near the angulardirection halfway between two adjacent RAs, the SINR at theMS (on the downlink) or at its serving RA (on the uplink) maybe substantially degraded,1 which we refer to as the “worst-casedirection” problem [23].

Motivations for the Comparative Study and Related Work:As detailed earlier, both the CoMP-CAS and the MR-FFR-DAShave their particular pros and cons, despite sharing a similarvirtual MIMO model. In general, the former scheme is morecomplex, and yet, its practically achievable performance maybe disappointing, as demonstrated in [8]–[11] and [17]–[19].Hence, in this paper, we aim for characterizing the cell-edgeperformance of the lower complexity MR-FFR-DAS in the con-text of the multicell uplink, which has not been disseminated

1This is because the cell-edge MSs of the same cell are operating in the samefrequency band in the MR-FFR-DAS.

in the open literature before. Additionally, we aim to providefurther insights into the question whether the MR-FFR-DASrelying on single-cell processing constitutes a promising techni-cal option in the scenario considered. Naturally, holistic cellularsystem design hinges on numerous technical aspects and targetspecifications; hence, it is a challenge to make “absolutely fair”comparisons between two system-level designs, and there isusually no definitive answer to the question of “which designis better.”

The existing BS-cooperation-aided CoMP reception tech-niques conceived for the multicell uplink typically rely on thephilosophies of either egoistic “interference cancellation” [16],[27], [28] or altruistic “knowledge sharing and data fusion”among BSs [6], [29], while both philosophies impose differentbackhaul traffic requirements. In our previous work [23], wehave shown that, in the downlink of the FFR-DAS, the intra-cell CCI imposed by the RAs may be mitigated by transmitpreprocessing dispensing with high-complexity multicell co-operation. As a beneficial result, the downlink throughput andcoverage quality in the cell-edge area of a multicell multiusernetwork may be significantly improved, where each BS simplyplays the role of the central signal processing (CSP) unit. Bycontrast, in the FFR-DAS uplink dispensing with multicellcooperation, the intracell CCI may be mitigated by single-cell multiuser detection (MUD) techniques. This is becausethe intracell CCI of the uplink is essentially constituted by themultiuser interference and the RAs are all connected to the BS,which is capable of carrying out centralized joint reception.However, it should be noted that, due to the geometry of theDAS—some RAs are close to the given cell-edge MSs andothers are far—the average receive SINRs recorded at differentRAs for a particular cell-edge MS are significantly different,which may cause the so-called “near–far problem.”

Novel Contributions: For the sake of characterizing theachievable cell-edge performance of both the CoMP-CAS andthe MR-FFR-DAS in the multicell uplink, we consider a setof four advanced MUD-based reception techniques, which aremore robust to the near–far problem than linear MUDs. Moreexplicitly, three noncooperative single-cell MUDs, namely,the classic minimum mean square error (MMSE)-based op-timal user ordering aided successive interference cancellation(MMSE-OSIC) [30], the “exhaustive brute-force search” basedmaximum-likelihood (ML) detection [31], and the probabilisticdata association (PDA) [32] scheme, are investigated for thesingle-cell processing that relies on neither BS cooperation norMRs. Furthermore, a low-backhaul-traffic knowledge sharingand data fusion based soft-combining PDA (SC-PDA) [6]scheme is conceived for both the CoMP-CAS and the MR-FFR-DAS. The novel contributions of this paper are summarizedas follows.

1) We demonstrate that the maximum achievable cooper-ative spatial diversity inherent in the MR-FFR-DAS islower than that in the CoMP-CAS, when each BS of bothsystems has the same number of antennas serving single-antenna MSs and only a single MR is invoked for eachcell-edge MS. Despite this, when assuming that the cell-edge MSs have a low transmission power and are located


in the close vicinity of RAs, even the FFR-DAS invokingno MRs may outperform the CoMP-CAS.

2) We show that, as an effective remedy to the aforemen-tioned worst-case direction problem, regardless of theMS positions, the SC-PDA-based receiver that relieson MRs is capable of providing a significant cooper-ative diversity gain compared with the noncooperativeMMSE-OSIC and PDA-based MUDs that invoke noMRs. Furthermore, the benefits of the SC-PDA-basedreceiver may be enhanced by carefully selecting the MRs’positions from an identified “reliable area.”

3) The QoS distribution of cell-edge MSs is visualized forthe MR-FFR-DAS, which demonstrates that, althoughpower control is an inefficient technique in interference-limited scenarios, it is more useful for improving thecell-edge MSs’ performance in the worst direction than inthe best direction of the MR-FFR-DAS considered. Thisinsight is valuable for improving the fairness among cell-edge MSs.

The rest of this paper is organized as follows. In Section II,we describe the multicell topology of both the CoMP-CASand MR-FFR-DAS regimes. In Section III, we detail the re-ceived signal models of both schemes. Then, in Section IV,the set of four MUD-based cooperative/noncooperative recep-tion schemes and the power control technique invoked aredescribed. The performance comparison results of the CoMP-CAS and of the MR-FFR-DAS are presented in Section V.Finally, our conclusions are offered in Section VI.

II. SYSTEM DESCRIPTION

A. Multicell Multiuser System Topology

1) MR-FFR-DAS Architecture: The MR-FFR-DAS architec-ture supporting a multicell multiuser operating scenario [23]consists of two tiers of 19 hexagonal cells, as shown in Fig. 1.The frequency partitioning strategy of the total available band-width F is characterized by Fc ∩ Fe = �, where Fc and Fe

represent the cell center’s frequency band and the cell edge’sfrequency band, respectively. Furthermore, Fe is divided intothree orthogonal frequency bands Fi, i ∈ {1, 2, 3}, which areexclusively used at the cell edge of each of the three adjacentcells. We have demonstrated that this regime is capable ofsufficiently reducing the intercell CCI in the FFR-DAS [23];hence, we can focus our attention on mitigating the CCI insidea single cell, as shown in Fig. 1. We consider a symmetricnetwork, where every cell has the same system configuration.Without any loss of generality, we focus our attention on cellB0, which is assumed to be at the origin of Fig. 1.

In the case of the MR-FFR-DAS arrangement in Fig. 1,we assume that Nr RAs are employed and a total of Nms

active MSs are roaming in the cell-edge area. Additionally, toincrease the attainable diversity gain, in each scheduling period,Nmr half-duplex MRs are invoked for supporting our FFR-DAS. Furthermore, a single omnidirectional antenna elementis employed both by each RA and by each MS. For the sake ofsimplicity, Nms = Nmr = Nt is assumed. The Nt MRs roam-ing in the cell-edge area are denoted by M e

k , k ∈ {1, . . . , Nt},

Fig. 1. Cellular topology of the MR-FFR-DAS architecture, where Nr = 6distributed antennas are employed and Nt = 6 MSs randomly roam in the cell-edge area.

which are identified by their polar coordinates, similarly to theactively communicating MSs, as shown in Table I.

2) CoMP-CAS Architecture: The CoMP-CAS architectureconsidered is shown in Fig. 2, where each single cell is dividedinto three 120◦ sectors Si, i ∈ {1, 2, 3} [33] and Nr collocatedantennas are employed at each BS [7]. We assume that theclassic frequency-division multiplexing, which is associatedwith Fi, i ∈ {1, 2, 3}, is used for the corresponding sectors Si.Hence, every set of three adjacent BSs constitutes a CoMPtransmission/reception area, as shown in Fig. 2, where B0 isassumed to be at the origin. We assume that there are Nt activeMSs in the CoMP area in Fig. 2 and all these MSs transmit at thesame frequency on the uplink. Additionally, each MS employsa single omnidirectional antenna element, while roaming in theCoMP area.

The BS-cooperation-aided cellular network in Fig. 2 is alsosymmetric. Hence, without any loss of generality, we assumethat the MS Zb

1 roams along the line B0C, i.e., in the directionbetween B0 and the center of the CoMP area, whereas theremaining MSs Zb

k, k ∈ {2, . . . , Nt} randomly roam across theentire B0-centered cell area according to a uniform distribution.Their polar coordinates are shown in Table I.

Hence, when observing the Nt active MSs, the multiplevirtual MIMO channel matrices of the cell-edge area transmis-sions taking place in our MR-FFR-DAS scheme in Fig. 1 andthe single virtual MIMO matrix of the conventional CoMP-CAS scheme2 shown in Fig. 2 have the same size of (Nr ×Nt) elements. Our MR-FFR-DAS scheme gleans cooperativediversity gain from the MRs, albeit this is achieved at the costof invoking a two-time-slot cooperation protocol. To elaboratea little further, the Nt active MSs transmit during the first timeslot, and the corresponding Nt MRs retransmit their receivedsignal in the second time slot. By contrast, the conventional

2A more detailed description of these virtual MIMO matrices is given inSection III.


TABLE ITOPOLOGY PARAMETERS FOR MR-FFR-DAS AND

COMP-CAS SCHEMES

Fig. 2. Topology of the CoMP-CAS composed of three adjacent BSs, whereNr = 6 collocated antennas are employed at each BS and Nt = 6 MSsrandomly roam in the cell-edge area.

CoMP scheme gleans its cooperative diversity gain from theadjacent two BSs.3

3) Digital Fiber Soliton Aided Backhaul: Until recently,the optical fiber backhaul has been assumed to be a perfectchannel, when transmitting low-rate data using on–off keying.However, when aiming for supporting gigabit transmissions,which is the ambitious goal of LTE-A standard [2], the high-ratefiber-based backhaul may suffer from the detrimental effects of

3In practice, for CoMP-aided systems, in addition to the time used by thetransmission between the BS and the MSs, there is an additional delay imposedby sharing data/channel information between the collaborative BSs. Morespecifically, in the CoMP-CAS uplink considered, we also need more than one,if not two, time slot to finish a single transmission: in the first time slot, the MSstransmit, and in the second time slot, the collaborative BSs transmit the decodeddata to each other for the sake of making a final decision. Therefore, in termsof the transmission time required, the schemes considered may be regardedidentical.

both dispersion and nonlinearity [34]. Since our MR-FFR-DASscheme still relies on centralized signal processing4 at the BS,where the signals are received from the RAs via optical fiberlinks, these signals may be contaminated both by the fiber’s dis-persion and by its nonlinearity. Fortunately, these degradationsmay be effectively mitigated by using the fiber soliton [35],[36]. To elaborate a little further, the fiber soliton techniqueof [35] and [36] is capable of improving both the linear andnonlinear distortions of the optical fiber backhaul. As a result,the optical pulse can propagate over the optical fiber with nodistortion, despite the interplay between the dispersive andnonlinear effects [36]. Fig. 3 shows a single optical fiber linkfrom an RA to the BS, where QPSK modulation is applied onthe uplink of the MR-FFR-DAS scheme. The signals receivedby the RAs from the wireless channel are first downconvertedto the baseband. Then, the I- and Q-streams are modulated byoptical pulses, and the resultant optical signaling pulses aretransmitted through the optical fiber.

III. RECEIVED SIGNAL MODELS OF THE UPLINK MOBILE

RELAY ASSISTED FRACTIONAL FREQUENCY REUSE

DISTRBUTED ANTENNA SYSTEM

As shown in Section II, both the MR-FFR-DAS in Fig. 1 andthe CoMP-CAS in Fig. 2 may be modeled relying on the virtualMIMO concept. For the sake of clarity, the signal transmissionprocess of the MR-FFR-DAS scheme considered is illustratedin Fig. 4. We can see from Figs. 2 and 4 that both the FFR-DASscheme using no MRs and the conventional CoMP-CAS maybe modeled as a multisource multidestination network havingdirect links only, where the multiple destinations are intercon-nected, and hence, they are capable of conducting collaborativesignal processing. Therefore, their wireless transmission partmay be modeled as a single (Nr ×Nt)-element virtual MIMOsystem. By contrast, the FFR-DAS scheme assisted by Nt half-duplex MRs may be modeled as a multiuser multirelay networkhaving both a direct link and a two-hop relay link for each MS.We assume that each MS is served by (possibly) multiple RAelements and a single selected MR, which is located betweenthe MS and the serving RA elements. Additionally, the servingMRs of the MSs transmitting their signals simultaneously areassumed to be sufficiently far from each other. Hence, when ob-serving the kth MR in Fig. 4, the interference from the jth MS,j �= k, may be ignored.5 As a result, the wireless transmissionpart of the MR-FFR-DAS scheme may be modeled as a pair of(Nr ×Nt)-element virtual MIMO subsystems (accounting forthe direct links from the MSs to the RAs and the links from theMRs to the RAs, respectively) and a subsystem with Nt parallelsingle-input–single-output links.

4The distributed RAs themselves do not have computing power, but onlycollect or radiate signals.

5Note that this assumption is indeed realistic upon invoking carefully de-signed MR selection, as detailed in Section V-C2. Since the MRs are single-antenna nodes and are distributed, it is infeasible for them to conduct jointdetection. Thus, the virtual MIMO system model is inappropriate for thetransmissions between the MSs and the MRs. By contrast, we do not have toimpose this assumption if more complex multiantenna MRs are used, because,in this case, the multiuser joint detection technique can be employed at eachMR for mitigating the impact of interference imposed by other MSs.


Fig. 3. System architecture of digital fiber optical link.

Fig. 4. Diagram of a DAS-based network with Nt MSs assisted by Nt MRs,which is modeled as a multiuser multirelay system. The signal received at Nr

RAs via the wireless channel is transmitted through the optical fiber to the BSthat plays the role of CSP.

A. Received Signal of a Single Optical Fiber Link

For the ith optical fiber link RAi → BS, i = 1, 2, . . . , Nr,the soliton technique of [36] may be applied for eliminatingboth the linear and nonlinear distortions. As a result, a near-perfect optical fiber backhaul may be created, where the opticalsignaling pulses are capable of propagating without distortion,as mentioned in Section II-A3. Hence, the phase rotationimposed by the optical fiber link on the modulated signal isnegligible, and the modulated signal’s amplitude is also main-tained, albeit, naturally, the modulated signal is contaminatedby the complex-valued additive white Gaussian noise (AWGN)nf ∼ CN (0, σ2

f ) at the BS receiver.As far as the wireless transmission part is concerned, let

us consider the direct links between the MSs and RAi as anexample. Then, the signal received at RAi from the wirelesschannel may be written as

ri =

Nt∑k=1

ζikhikxk + nw (1)

where xk, ζik, hik, and nw ∼ CN (0, σ2w), k = 1, 2, . . . , Nt,

i = 1, 2, . . . , Nr, represent the signal transmitted from MSk,the path loss between MSk and RAi, the small-scale Rayleighfading coefficients between MSk and RAi, and the AWGN

corresponding to the signal ri received at RAi, respectively.6

Then, the received signal of the ith optical fiber link at theBS may be written as yi = χri + nf , which is expressed moreexplicitly as

yi =

Nt∑k=1

χζik︸︷︷︸gik

hikxk + χnw + nf︸︷︷︸ni

(2)

where χ is the power-scaling factor invoked for ensuringthat the peak power of the optical signaling pulse obeys thefundamental soliton requirement of [36]. Hence, gik = χζikrepresents the equivalent MSk-RAi-BS link’s large-scale atten-uation. Finally, ni = χnw + nf denotes the equivalent receivernoise jointly induced by the optical fiber and the wirelesschannel. It is also worth noting that, for each MSk, an obviousnear–far effect is observed at the RAs. More explicitly, for eachMSk, the SINRs at the RAs may differ significantly, since MSk

is mainly served by its nearest RA.Additionally, when MRs are invoked, the signal received at

RAi from the MRs may be characterized in a similar fashion to(1) and (2), but with smaller path loss, different power-scalingfactor, and different channel fading coefficients.

B. Received Signal Model of the Virtual MIMO Observedat the BS

The channel state information (CSI) of all MSk-RAi-BSlinks is assumed to be perfectly known at the central BS. Again,let us consider the direct links from the Nt MSs to the BS as anexample. Based on (2), the signal vector received at the BS onthe idealized synchronous uplink may be written as

y = Hx+ n (3)

where y ∈ CNr×1, H ∈ CNr×Nt , and n ∈ CNr×1 denote thereceived signal vector, the channel matrix containing the per-fect CSI, and the noise vector, respectively. Furthermore, x =[x1, x2, . . . , xNt

]T represents the symbol vector transmittedfrom the Nt MSs, whereas n = [n1, n2, . . . , nNr

]T stands forthe circularly symmetric complex Gaussian effective noise vec-tor at the BS, and its element ni ∼ CN (0, σ2) was defined in

6We assume that the noise variance of each direct wireless link remains equalto σ2

w .


(2). Still referring to (3), we have H = [hT1 ,h

T2 , . . . ,h

TNr

]T

,

where hi = [hi1, hi2, . . . , hiNt] ∈ C1×Nt represents the chan-

nel vector between all the Nt MSs and the BS via a particularRAi, and hik is defined as gikhik according to (2). The entriesofH account for both the path loss and the small-scale Rayleighfading of the wireless channel, as well as the impact of theoptical fiber link, i = 1, 2, . . . , Nr, k = 1, 2, . . . , Nt.

The received signal model introduced earlier in (2) and (3)can be extended in a straightforward manner to both the secondtime slot of the MR-FFR-DAS scheme in Fig. 1 and to theconventional CoMP-CAS in Fig. 2. However, in the CoMP-CAS, we have ni = nw if a wireless backhaul is employedamong the collaborative BSs, and typically, a higher path lossζik is imposed on the cell-edge MSs compared with the DASscheme [37].

C. Correlation Between the Channel Coefficients ofMSk-RAi-BS and MRk-RAi-BS Links

We assume that the decode-and-forward (DF) relaying pro-tocol is invoked in the multiuser MR-FFR-DAS scheme inFig. 4. Since the MRs are geographically distributed, jointdetection/decoding at the MRs is infeasible. Therefore, in thefirst time slot, we have Nt single-input–single-output linksand an (Nr ×Nt)-element virtual MIMO subsystem that relieson the signal model of (3). More specifically, as shown inFig. 4, the Nt single-input–single-output links account for thetransmissions from the Nt MSs to their selected Nt MRs, andthe virtual MIMO system characterizes the direct transmissionsfrom Nt MSs to the BS via the Nr RAs. Furthermore, in thesecond time slot, the MSs remain silent when the Nt MRsretransmit the decoded information to the BS, again, via the Nr

RAs. Hence, the channel model of the second time slot may bealso regarded as an (Nr ×Nt)-element virtual MIMO model,similar to (3).

Similar to the virtual MIMO channel H of (3), whichcharacterizes all the MSk-RAi-BS links, let us instantiate thechannel matrix representing all the MRk-RAi-BS links as HR,where we have HR = [hT

1 , hT2 , . . . , h

TNr

]T ∈ CNr×Nt . Moreexplicitly, during the second time slot, for the MRs retrans-mitting their signals received in the first time slot, we havehi = [hi1, hi2, . . . , hiNt

] ∈ C1×Nt , i ∈ [1, . . . , Nr], which rep-resents the channel vector from all the Nt MRs to the BS via aparticular RAi.

When the MSk is close to the selected MRk and far fromthe other MRs,7 the MSk-RA-BS link and the MSk-MRk-RA-BS link may exhibit a high envelope correlation, and theinterference imposed by MSk on the rest of the MRs maybe negligible. Additionally, to gain fundamental insights, inthis scenario, it is reasonable to assume that the selected MRk

employs perfect DF relaying, implying that the source signal ofthe kth MS is perfectly decoded at the kth MR [39], although inpractice, we might still observe some degree of decoding error

7The selected MRk exhibits the best performance among all MRS for theMSk-MR-RAi link, and it is not necessarily the spatially nearest MR to MSkor to the corresponding RAi [38].

propagation at the MR. As a result, the single-input–single-output MSk-MRk link becomes lossless, and only the envelopecorrelation between the MRk-RAi-BS channels HR and thedirect MSk-RAi-BS links’ channel H is relevant. This envelopecorrelation is given by [40]

ρik=E(|hik||hik|)− E(|hik|)E(|hik|)√(

E(|hik|2)− [E(|hik|)]2)(

E(|hik|2)−[E(|hik|)]2)

(4)

where k ∈ [1, . . . , Nt], and i ∈ [1, . . . , Nr]. For ρik = 0, theMSk-RAi-BS link in the first time slot and its correspondingMRk-RAi-BS link in the second time slot are regarded to beuncorrelated.

Similarly, when the selected relay MRk is close to RAi, weassume that the signal transmitted by MRk is perfectly receivedat RAi and that the interference imposed by MRk on the restof the RAs is negligible. Hence, only the envelope correlationbetween the MSk-MRk link and the direct MSk-RAi link isrelevant, which may be characterized in a similar fashion to (4).

IV. CENTRAL SIGNAL PROCESSING AT THE BASE STATION

The philosophy of the DAS architecture relies on invokingRAs for transmitting (on the downlink) and receiving (on theuplink) the signals, which facilitates the centralized processingof the virtual MIMO signals at the BS, since the BS can affordto apply more complex MUD techniques. For a forward errorcorrection (FEC)-coded system, if a soft MUD is invoked, thelog-likelihood ratio (LLR) of each coded bit is calculated basedon the signal vectors received at the BS. Then, the LLR ofeach bit is subjected to soft decoding, since soft decoding iscapable of achieving a better performance than hard decoding.For the sake of maintaining a low computational complexity,in this paper, we opt for an open-loop soft receiver dispensingwith iterations between the MUD and the FEC decoder. Beforeintroducing our powerful PDA-based soft MUD that is capableof achieving an attractive performance at the expense of amoderate computational complexity, let us first introduce theoptimal ML-based soft MUD and the classic MMSE-OSIC-based hard-decision MUD as our benchmarkers.

A. Optimal ML-Based Soft MUD

For an FEC-coded virtual MIMO system, the soft ML-MUDis the optimum detector, albeit it imposes a potentially excessivecomplexity. The soft ML-MUD calculates the LLR for the nthbit bk,n of the kth MS according to

L(bk,n) = logP (y|bk,n = 1)P (y|bk,n = 0)

. (5)

The max-log approximation may be invoked for reducing thecomplexity at a negligible performance degradation. Hence, the


LLR in (5) may be reformulated as [31]

L(bk,n)

= log

⎛⎜⎝ ∑

x1∈A,...,xk∈A(1)n ,...,xNt∈A

exp

(−‖y−Hx‖2

σ2

)⎞⎟⎠

− log

⎛⎜⎝ ∑

x1∈A,...,xk∈A(0)n ,...,xNt∈A

exp

(−‖y−Hx‖2

σ2

)⎞⎟⎠

≈ 1σ2

(min

x1∈A,...,xk∈A(0)n ,...,xNt∈A

‖y−Hx‖2)

− 1σ2

(min

x1∈A,...,xk∈A(1)n ,...,xNt∈A

‖y −Hx‖2)

(6)

where A denotes the constellation alphabet, and A(b)n represents

the set of constellation symbols whose nth bit is equal to b ∈{0, 1}. Hence, the size of A(b)

n is half of that of A. More explic-itly, for calculating L(bk,n), each of the two min operators in(6) calculates (1/2)|A|Nt Euclidean distances (EDs) and findstheir minimum. Therefore, in total, |A|Nt EDs are calculated,which represents a brute-force search over all possible valuesof the symbol vector x. In other words, the size of the solutionspace increases exponentially with the number of MSs (or MRs)in the MR-FFR-DAS scheme considered.

B. MMSE-OSIC-Based Hard-Decision MUD

A classic reduced-complexity suboptimum MUD solution isconstituted by the MMSE-OSIC detector, which is capable ofstriking a tradeoff between reducing the intracell CCI imposedby the other cochannel MSs of the same cell and mitigating theimpact of the Gaussian background noise [41] in the MR-FFR-DAS scheme considered. Relying on (3), the basic principleof the OSIC operations may be described as follows. Amongall the elements of x, the specific symbol element that hasthe highest receive signal-to-noise ratio (SNR), for example,xk , was first decoded and remodulated. Then, the impact ofthe corresponding regenerated transmit signal, which is theproduct of xk and the kth column vector of H, is subtractedfrom the received signal vector y. Subsequently, this process isrepeated in the next interference cancellation stage relying onthe residual received signal vector. Theoretically, the MMSE-OSIC detector is capable of approaching the MIMO capacity[41]. Unfortunately, in practice, this optimality is undermined,since the MMSE-OSIC detector is prone to interlayer errorpropagation in the presence of decision errors at each layer [41].

C. PDA-Based Soft MUD

As noted in Section IV-A, the optimal ML-based MUD hasa computational complexity that increases exponentially withthe number of MSs simultaneously served. Hence, it may notbe invoked in practical cellular systems having a high numberof MSs. As an attractive low-complexity design alternative, the

PDA algorithm that is capable of generating bit LLRs for theconcatenated channel decoder without a brute-force search maybe applied in our MR-FFR-DAS scheme. In the uplink scenarioconsidered, we assume that the CSI H is unknown at thetransmitters of the MSs, but it can be accurately estimated at thereceiver of the BS. Furthermore, for the sake of convenience,Nt = Nr is assumed, and the decorrelated signal model [6] isadopted. Hence, (3) may be reformulated as

y = x+ n = xkek +∑j �=k

xjej + n

︸︷︷︸vk

(7)

where y = (HHH)−1HHy, n is a colored Gaussian noisevector with zero mean and covariance matrix of N0(H

HH)−1,ek is a column vector with 1 in the kth position and 0 elsewhere,whereasvk denotes the interference-plus-noise term for symbolxk, k ∈ [1, . . . , Nt]. For each symbol xk, we have a probabilityvector pk, whose mth element P(xk = am|y) quantifies thecurrent estimate of the nominal a posteriori probability (APP)8

of having xk = am,m ∈ [1, . . . ,M ] upon receiving y, wheream represents the mth element of the modulation constellationalphabet A.

The basic idea of the PDA algorithm is to iteratively approx-imate the complex random vector vk that obeys the multimodalGaussian mixture distribution as a single multivariate coloredGaussian distributed random vector having an iteratively up-dated mean, covariance, and pseudocovariance9 given by

E(vk) =∑j �=k

E(xj)ej (8)

C(vk) =∑j �=k

C(xj)ejeTj +N0(H

HH)−1 (9)

C(vk) =∑j �=k

C(xj)ejeTj (10)

respectively, where we have

E(xj) =

M∑m=1

amP(xj = am|y) (11)

C(xj) =

M∑m=1

[am − E(xj)] [am − E(xj)]∗ P(xj = am|y)

(12)

C(xj) =

M∑m=1

[am − E(xj)]2 P(xj = am|y). (13)

For estimating P(xk = am|y), it is initialized as 1/M basedon the uniform distribution, and then, it is updated at eachiteration of the PDA algorithm, which is a process of gradually

8As shown in [32], since an approximate form of Bayes’ theorem is typicallyinvoked in the PDA algorithm, this nominal APP is essentially the normalizedsymbol likelihood, rather than the true APP.

9The pseudocovariance is necessary for characterizing complex randomvector that is improper [42]–[44].


reducing the decision uncertainty concerning the event xk =am|y. Let

w(k)m =y − a(k)m ek − E(vk) (14)

ϕm(xk)Δ= exp

⎛⎜⎝−

⎛⎝�

(w

(k)m

) (w

(k)m

)⎞⎠T

Λ−1k

⎛⎝�

(w

(k)m

) (w

(k)m

)⎞⎠⎞⎟⎠(15)

where the composite covariance matrix of vk is given by

ΛkΔ=

(� [C(vk) + C(vk)] − [C(vk)− C(vk)] [C(vk) + C(vk)] � [C(vk)− C(vk)]

)(16)

and akm indicates that am is assigned to xk; whereas �(·)and (·) represent the real and imaginary parts of a complexvariable, respectively.

Since no external source of the a priori probability P (xk =am) is available, the decision probability of the event xk =am|y is approximated as

P(xk = am|y) = p(y|xk = am)P (xk = am)∑Mm=1 p(y|xk = am)P (xk = am)

≈ ϕm(xk)∑Mm=1 ϕm(xk)

. (17)

Hence, an updated value of P(xk = am|y) has been obtained.Based on this updated value, the preceding decision probabilityestimation process is repeated until P(xk = am|y) has con-verged for all values of k and m. Then, the bit LLR L(bk,n)delivered to the channel decoder is given by [32]

L(bk,n|y) = ln

∑∀am∈A(1)

nP(xk = am|y)∑

∀am∈A(0)n

P(xk = am|y) . (18)

For further details on the PDA-aided MUD, see [6] and [32].

D. Combining the Soft Information of the MSs and the MRsat the BS

When using channel codes that support soft-input–soft-output decoding (such as convolutional codes, turbo codes, andLDPC codes), the outputs of the channel decoder are also bitLLRs. To achieve a higher diversity gain in the MR-FFR-DASscheme considered, the soft-decision information gleaned fromthe MSs and the MRs may be combined in a manner similarto that proposed for the BS-cooperation-aided CoMP-CAS of[6]. However, channel codes were not considered in [6]; hence,soft information combining was implemented in the probabilitydomain in [6]. By contrast, since each active MS is assisted byan appropriately selected MR employing the DF protocol in ourchannel-coded MR-FFR-DAS scheme, the channel decoders ofthe MRs also generate bit LLRs, which may be forwarded to theRAs. Therefore, the soft information combining in our channel-coded MR-FFR-DAS scheme has to be implemented in theLLR domain.

More specifically, if no channel codes are employed, thedecision probabilities P(xk = am|yMS) calculated at the BS

based on the direct transmission during the first time slot andP(xk = am|yMR) calculated relying on the MR-aided trans-mission in the second time slot are combined as [6]

P(xk = am|yC) =P(xk = am|yMS)P(xk = am|yMR)∑m P(xk = am|yMS)P(xk = am|yMR)

(19)for k ∈ [1, . . . , Nt]. Then, the BS’s MUD makes a decision foreach transmitted symbol xk, yielding xk = am′ , where we have

m′ = arg maxm=1,2,...,M

{P(xk = am|yC)} . (20)

By contrast, when soft-decoded channel codes are invoked, thesoft information generated by the MUD of the BS from thedirect transmission during the first time slot and from the MR-aided relay transmission during the second time slot may besimply combined as

LC(bk,n) = LMS(bk,n) + LMR(bk,n). (21)

Then, LC(bk,n) is fed into the channel decoder to generate thefinal decoding results.

E. Power Control in the Multiuser Uplink Scenario

Perfect CSI is typically unavailable in practice, whereas hav-ing an imperfect CSI leads to a degraded performance. Hence,we introduce power control for improving the SINR of the MSsroaming in the cell-edge area. We use the normalized signal-to-interference ratio (SIR)-based model for investigating the effectof the multiuser interference in the uplink, since the AWGNsof both the wireless and optical fiber links are moderate. Inother words, in the multiuser multicell scenario considered, itis the effect of the interference, rather than the AWGN, thatdominates the attainable performance. This is the so-calledinterference-limited scenario. More explicitly, in interference-limited systems, we have SINR ≈ SIR, and we record the SIRfor all MSs randomly roaming across the cell-edge area. Usingno power control, we adopt the theoretical SIR model that takesno account of fading of any kind and is solely determined by thepath loss. This simplified SIR model is expressed as [45, Ch. 7]

SIRi(dB) = 37.6 log10didw

(22)

SIRMS(dB) = − 10 log10

⎡⎣ ∑i=1,...,Nt−1

10−SIRi(dB)

10

⎤⎦ (23)

where SIRi represents the SIR experienced by the wanted MSwhen there is only a single interferer having the index i. Morespecifically, dw is the distance between the wanted MS andits serving RA, whereas di is the distance between the singleinterfering MS and the RA that serves the wanted MS. Weassume a 37.6-dB/decade inverse power path loss law. For themultiuser scenario considered in Fig. 1, we define SIRMS as theSIR recorded at its respective serving RA for any MS roamingacross the cell-edge area. Since a specific MS suffers from thecontaminating effects of (Nt − 1) interferers, SIRMS can bewritten in decibels as (23).


The geographic SIR distribution of the cell-edge area MSsis based on (23), where the specific MSs suffering from alower SIR will be assisted by increasing their transmit power.When applying power control, the SIR of the wanted MScontaminated by the ith interfering MS may be written as

SIR′i(dB) = 37.6 log10

didw

+ PwdB − P i

dB (24)

where di and dw were defined in (22). Furthermore, PwdB =

log10 PTw and P i

dB = log10 PTi represent the transmit power of

the wanted MS and that of the interfering users in decibels,respectively. More explicitly, when the MS roaming in the cell-edge area suffers from a lower SIR, the transmit power Pw

dB

or (P idB) will be increased (or decreased) in the interest of

maintaining the target SIR of this MS. Correspondingly, theinterfering MS may also suffer from a lower SIR, and hence,its transmit power P i

dB will be increased, which will, in turn,increase the interference imposed on the other MSs.

The expression of the SIR recorded in the presence of powercontrol for our multiuser scenario remains similar to that with-out power control. Hence, based on (23) and (24), we have

SIR′MS(dB) = −10 log10

⎡⎣ ∑i=1,...,Nt−1

10−SIR′

i(dB)

10

⎤⎦ . (25)

V. PERFORMANCE EVALUATION

Here, both the uplink BER and the effective throughput ofthe conventional noncooperative CAS, of the BS-cooperation-aided CoMP-CAS (illustrated in Fig. 2), of the noncooperativeFFR-DAS dispensing with MRs, and of the MR-FFR-DAS (il-lustrated in Fig. 1) are characterized. The simulation parametersare summarized in Table II, where d0 is the distance betweenany transmitter-and-receiver pair in kilometers. We define thecell-edge region as the area outside the radius of r = 0.5R;hence, by definition, the cell-center area is within the radiusof r = 0.5R, as shown in Fig. 1. The RAs in the cell-edgearea are assumed to be located at de = 0.7R. Four types ofMUDs, i.e., the high-complexity noncooperative soft ML, thenoncooperative MMSE-OSIC, the noncooperative PDA, andthe cooperation-based SC-PDA, are compared in the contextof both the CAS-based and FFR-DAS-based architectures, asshown in Figs. 5–9. More specifically, as a benchmarker, theclassic noncooperative MMSE-OSIC-based MUD consideredgenerates hard-decision information for the concatenated chan-nel decoder. Hence, the BER performance of the MMSE-OSICreceiver remains poor in both the CAS-based and FFR-DAS-based systems. By contrast, the soft-decision-based MUDs (i.e.,the soft ML, the PDA, and the SC-PDA) achieve a superior BERperformance, despite their increased complexity.

A. Calculation of the Effective Throughput

We define the effective throughput in terms of bits/second/hertz as follows:

Ceff = Craw × (1 − BER)Lp (26)

TABLE IISIMULATION PARAMETERS FOR THE UPLINK SCENARIOS CONSIDERED,

WHICH RELY ON THE URBAN MACRO PROPAGATION MODEL

OF THE 3GPP-LTE STANDARD [46]

where the packet length Lp is set to 1024 bits in our eval-uations.10 We assume that Nt MSs transmit over the wholechannel bandwidth simultaneously; hence, the raw throughputCraw is given by

Craw =Nt ×Rc ×Mb ×Rs ×Nsc

B. (27)

The specific parameter values invoked for calculating Craw aregiven in Table II.

B. BER Performance of the Conventional CAS and theBS-Cooperation-Aided CoMP-CAS

The BER performance of both the conventional CAS and theBS-cooperation-aided CoMP-CAS schemes is characterized inFig. 5, which is obtained by considering Nt = 6 cochannelMSs roaming randomly across a single cell and across theentire collaboration area composed of three sectors of three

10A packet loss event happens whenever any of the bits contained inthis packet is erroneously decoded at the receiver. The packet length doesnot change the general insights and conclusions drawn from our effectivethroughput comparisons.


Fig. 5. BER performance of the conventional CAS architecture using thenoncooperative MMSE-OSIC, PDA, and soft ML based MUDs and of theBS-cooperation-aided CoMP-CAS architecture employing the SC-PDA-basedMUD.

Fig. 6. BER performance of the MSs located at the θbest direction in the cell-edge area of the noncooperative FFR-DAS and the MR-FFR-DAS schemes. Nopower control is used, and a fixed MS/MR transmission power Pt = 20 dBmis assumed. The noncooperative high-complexity soft ML, the MMSE-OSIC,and the PDA based MUDs are invoked when MRs are not used, whereas theSC-PDA is used when MSs are assisted by the MRs. The MRs are selectedaccording to the simple “close-to-MS” strategy.

adjacent BSs, respectively, as shown in Fig. 2. Furthermore,Nr = 6 omnidirectional antennas were applied at each of theBSs. Hence, in each cell, there are six BS antennas servingeach sector, and for the CoMP-CAS, in total, there are 18 BSantennas serving the three collaborative sectors. As a result, theCoMP-CAS scheme relies on an equivalent (18 × 6)-elementvirtual MIMO system model, if perfect BS cooperation11 isassumed. When the MS/MR transmit power Pt is increasedfrom 20 to 30 dBm, the BERs of all MUDs considered areimproved. However, this BER improvement becomes moresubstantial when the MSs are close to the BS, as indicated by

11In perfect BS cooperation, the information to be shared among the collab-orative BSs is received at each BS without transmission error.

Fig. 7. BER performance of the MSs located at the θworst direction in the cell-edge area of the noncooperative FFR-DAS and the MR-FFR-DAS schemes.The other configurations are the same as those in Fig. 6.

smaller values of d/R. By contrast, when the MSs are roamingin the cell-edge area, as indicated by larger values of d/R, evenan increased transmit power Pt fails to result in a sufficientlycompetitive BER for the MS of interest. This is because thepath loss is rather high and the SINR is low in the cell-edgearea.

It is worth noting that the SC-PDA-based MUD is adoptedfor the CoMP-CAS scheme, which requires the cooperation ofthree adjacent BSs. Observe from Fig. 5 that the cooperationdiversity gain attained by the SC-PDA-based MUD becomesmore significant when the transmit power of each MS is ashigh as Pt = 30 dBm. By contrast, when the MSs transmit ata lower power of Pt = 20 dBm, the SC-PDA-based MUD has asimilarly poor BER performance to that of the noncooperativesingle-cell PDA-based MUD, which operates without exchang-ing soft information among the adjacent BSs. The reasonsbehind these observations are as follows. When the MSs roamclose to their own anchor BS, but quite far from other adjacentBSs, the cooperation gain remains rather limited, since thepath loss experienced by the MS of interest with regard to theadjacent BSs is high. This leads to inefficient cooperative BSprocessing. On the other hand, when the MSs roam close to thecell edge, a sufficiently high SINR may only be guaranteed for ahigh transmit power, since the path loss of the MSs with regardboth to their anchor BS and to any of the adjacent BSs remainshigh.

C. Performance of the Noncooperative FFR-DAS and theMR-FFR-DAS in the Uplink

Similarly, we consider Nt = 6 cell-edge MSs and Nr = 6RAs in the noncooperative FFR-DAS, whereas Nt = 6 addi-tional MRs are also invoked in the MR-FFR-DAS scheme, asillustrated in Figs. 1 and 4. On one hand, when no MRs areinvoked, a particular MS’s uplink signal received at the BS iscontaminated by the other cochannel MSs transmitting withinthe single time slot of the noncooperative FFR-DAS scheme. In


Fig. 8. Impact of the improved MR selection strategy on the (a) BER and the (b) effective throughput of the MSs located at the θbest direction in the cell-edgearea of the MR-FFR-DAS scheme. The “SC-PDA, ρ = 0” curves are obtained by selecting the MRs according to the simple close-to-MS strategy, which servesas a benchmark of the improved strategy that selects MRs only from the reliable area. The remaining configurations are the same as those in Fig. 6.

Fig. 9. Impact of the improved MR selection strategy on the (a) BER and the (b) effective throughput of the MSs located at the θworst direction in the cell-edgearea of the MR-FFR-DAS scheme. The remaining configurations are the same as those in Fig. 8.

this scenario, the performance of three noncooperative MUDs,namely, of the high-complexity soft ML, of the MMSE-OSIC,and of the PDA, are numerically evaluated. On the other hand,when Nt = 6 MRs are employed, the signal relayed by aparticular MR is also contaminated by the other cochannel MRstransmitting in the second time slot. In this context, an SC-PDA-based MUD capable of exploiting the user cooperationgain gleaned from MRs is examined. In contrast to the con-ventional BS-cooperation-aided CoMP-CAS scheme, the MR-FFR-DAS scheme attains a beneficial cooperation diversitygain with the assistance of MRs, which is facilitated by usinga cooperation time slot as well. Furthermore, this cooperationdiversity gain highly depends on the locations of the MSs.Among them, two specific directions, as represented by θbestand θworst, are of particular interest, since they serve as thebounds for the general case.

1) Cell-Edge Area Performance Without Power Control:In Fig. 6, we investigate the BER performance of both thenoncooperative FFR-DAS and the MR-FFR-DAS schemes,where the MSs are located at the θbest direction in the cell-edge area, as shown in Fig. 1. No power control techniqueis used, and a fixed MS/MR transmission power of Pt =20 dBm is assumed. Hence, for a particular MS, upon assuminga constant noise power and interfering MSs’ locations, theSINR attained at the BS is mainly determined by the dis-tance from this MS to the RA, as demonstrated in (23). Wecan see from Fig. 6 that, when the MSs roam close to theRAs, which are at the location of d/R = 0.7, a high SINRmay be obtained. Therefore, even the noncooperative hard-decision MMSE-OSIC-based MUD is capable of achievinga low BER. However, when the MSs roam far away fromthe RAs, namely, toward either d/R = 0.5 or d/R = 1, the


achievable SINR gradually becomes lower. As a result, boththe noncooperative MMSE-OSIC and the noncooperative PDAfail to achieve an appealing BER in the noncooperative FFR-DAS. By contrast, when each MS is assisted by an MR, whichis selected according to the simple strategy that the one selectedhas to be reasonably close to the MS for the sake of attaining adiversity gain in the MR-FFR-DAS, the SC-PDA-based MUDis invoked. Since the SC-PDA is capable of efficiently com-bining the soft information gleaned from the MSs and MRs, itattains a BER typically lower than 10−5 in most locations of thecell-edge area, as defined by d/R ∈ [0.5, 0.9]. Note, however,that when the MSs are very close to the cell boundary, asindicated by d/R → 1, all of the schemes considered exhibitpoor BER performance.

Additionally, the performance of the MSs at the θworst di-rection is characterized in Fig. 7. In contrast to Fig. 6, weobserve that the BER performance of even the SC-PDA-basedMUD is also dramatically degraded. This is because, in theθworst direction, MSk roams in the area where the desired signalreceived at RAk may in fact be weaker than the interferenceimposed by other cochannel MSs.

Therefore, it is straightforward to infer that the BER perfor-mance across the entire cell-edge area is between that of thebest case angle θbest, as characterized in Fig. 6, and that of theworst-case scenario θworst, as shown in Fig. 7.

Furthermore, the dotted curves shown in Figs. 6 and 7 quan-tify the impact of the correlation between the channel coeffi-cients of MSk-RAk-BS and MRk-RAk-BS, as defined in (4),when the MR is selected in the vicinity of the MS. The rangeof the correlation coefficient ρ examined is between 0 and 1.More explicitly, when the direct MSk-RAk-BS link and thecorresponding MRk-RAk-BS link are uncorrelated, i.e., wehave ρ = 0, the SC-PDA-based MUD achieves a diversity gaincontributed by a pair of uncorrelated channel matrices H andHR. By contrast, when the correlation is approaching ρ = 1,the diversity tends to be completely eroded; hence, in this case,the SC-PDA has a similar BER performance to that of thenoncooperative PDA.

2) Improved MR Selection in the Reliable Area: For a partic-ular cell-edge MS and its corresponding RA, the MR selectedhas a location between them. In Figs. 6 and 7, when a smallρ is required for obtaining an increased diversity gain, thespacing between the MR selected and the cell-edge MS has tobe sufficiently large, albeit the MR is still selected in the vicinityof the cell-edge MS considered12 rather than in the vicinityof the RA or the BS. More explicitly, the distance betweenthe MR and the cell-edge MS should remain sufficiently largeso that the correlation between the direct link and the secondhop of the relayed link remains low. In this case, the SC-PDA-based MUD effectively combines the soft informationgleaned from the signals transmitted by both the MSs andthe MRs, and a substantially improved BER may be achieved

12The “vicinity” of a cell-edge MS may be defined as a half-circle region,which is centered at the cell-edge MS and located between this cell-edge MSand its nearest RA. When both the cell-edge MSs and the MRs are uniformlyrandomly distributed in the cell, it is unlikely that the MRs selected for differentcell-edge MSs will be very close to each other.

Fig. 10. Average SIR experienced by cell-edge MSs of the MR-FFR-DASscheme operating both with and without power control in the θbest and θworst

directions.

by the SC-PDA compared with both the noncooperative PDAand the noncooperative MMSE-OSIC. However, there is stillan area where the BER cannot be significantly reduced evenwhen invoking the SC-PDA. This is encountered when the MSroams very far away from the RA, as indicated by d/R → 1.More explicitly, when the MR is selected to be close to theMS that approaches the cell-edge boundary, both of them maybe too far from the RA. As a result, the selected MRs alsosuffer from a high level of path loss, and the SC-PDA remainsunable to effectively reduce the BER. Therefore, as far as fixedRAs are considered, for the sake of maintaining an adequateperformance, the cell-edge MSs should be neither too close tothe BS nor too close to the cell-edge boundary.

As a solution to this predicament, we may always select theMR from the reliable area so that the MR selected is not toofar from the RA, regardless of the ordinary cell-edge MS or theMS that is very close to the cell-edge boundary, as illustratedin Figs. 8 and 9 for both the θbest and θworst directions,respectively. Theoretically, the optimal position of the selectedMR under the DF protocol should strike an attractive tradeoffbetween being close to the cell-edge MS and being not too farfrom the RA for the sake of having a good performance at boththe first and second hops. As demonstrated in Fig. 8(a), as faras the SC-PDA is concerned, compared with the simple close-to-MS MR selection strategy described in Section V-C1, theoriginal high BER experienced by the MSs when they roam faraway from the RAs but remain near the θbest direction maybe improved by using this improved MR selection strategy.This is evidenced by the curve marked by the hollow stars inFig. 8(a). On the other hand, Fig. 8(b) shows that, for cell-edgeMSs that are near to the RAs, the SC-PDA-based MR-FFR-DAS scheme, which requires two time slots for completingeach MS’s transmission, has a lower effective throughput thanthe noncooperative FFR-DAS that uses only a single timeslot for completing the same task. However, when the MSsapproach the cell-edge boundary, the MR-FFR-DAS providesa better effective throughput. This observation confirms thethroughput-erosion impact of the extra time slot imposed on


Fig. 11. BER and effective throughput of the noncooperative FFR-DAS and the MR-FFR-DAS schemes operating with power control in the cell-edge area whenapplying the noncooperative-PDA-based and the SC-PDA-based MUD techniques, respectively. Both the θbest and θworst directions are evaluated.

the effective throughput in cooperative systems. It also demon-strates that introducing MRs is indeed particularly beneficialfor the MSs roaming close to the cell-edge boundary in termsof both the BER and the effective throughput. Additionally, thebenefits of the improved MR selection strategy are corroboratedby the effective throughput results in Fig. 8(b).

Similarly, we can see from Fig. 9(a) that the improved MRselection strategy is also beneficial for the θworst direction interms of BER. However, the performance gain is not as high asthat of the θbest direction. This is because the selected MRsemploying the DF protocol have to be closer to the sourcethan to the destination, and compared with the scenario of theθbest direction having the same value of d/R, the cell-edgeMSs are farther away from the RAs in the θworst direction.As a result, the selected MRs are also farther from the RAs,and hence suffering higher path loss. Additionally, when eachθworst direction of a cell has a cell-edge MS, each selectedMR encounters an increased interference impact comparedwith the scenario of the θbest direction. However, in terms ofthe effective throughput at the cell edge, it is observed fromFig. 9(b) that, apart from the high-complexity soft ML-basedMUD, all the noncooperative MUDs effectively fail. This isbecause, in the θworst direction, the BER performance of thenoncooperative MMSE-OSIC and of the noncooperative PDAis so poor that hardly any packet can be successfully decoded.Additionally, in certain parts of the cell-edge area, the SC-PDA-based MR-FFR-DAS schemes indeed achieve a significantlyhigher effective throughput than the noncooperative FFR-DASthat employs either the MMSE-OSIC or the PDA. However,their achievable effective throughput still remains much lowerthan that of the noncooperative FFR-DAS employing the softML-based MUD, particularly when the cell-edge MSs approachthe cell-edge boundary.

3) Power Control in θbest and θworst Directions: We haveobserved from Figs. 8 and 9 that, although the BER of theMSs in the FFR-DAS-based schemes has been substantiallyimproved in most of the cell-edge area with the aid of the SC-PDA-based MUD invoking MRs, the BER of MSs roaming

Fig. 12. Number of simultaneously supported MSs satisfying the QoS ofSIR > 15 dB.

very close to the cell-edge boundary (e.g., at the location whered/R = 1) remains higher than 10−3, even when activating anMR in the reliable area of the θbest scenario. As a result,in terms of the effective throughput calculated relying on thepacket error rate while assuming the packet length of 1024 bits,the cell-edge MSs are still poorly supported by the FFR-DAS-based schemes. As a further remedy, we employ power controlfor improving the SIR for the cell-edge MSs roaming far awayfrom the RAs.

The average SIR experienced by cell-edge MSs at the θbestand θworst directions is recorded both in the absence and inthe presence of power control, as shown in Fig. 10. For thescenario of the θbest direction, we can see that, when theMSs are roaming close to the RAs in the cell-edge area, i.e.,d/R → 0.7, a high average SIR of SIRMS ≥ 20 dB is main-tained without increasing the transmit power Pt. By contrast,when power control is used, a similarly high average SIRof SIR′

MS ≥ 20 dB is also experienced by the MSs roamingclose to the RAs. However, in this context, the average SIR


Fig. 13. SC-PDA-aided CoMP-CAS versus the SC-PDA-aided MR-FFR-DAS in terms of their (a) BER and (b) effective throughput in the cell-edge area.

value obtained is in general slightly lower than that recordedin the absence of power control. This is because, in the systemwhere multiple users share the same frequency bandwidthsimultaneously, power control is an inefficient technique froma sum-capacity perspective. To elaborate a little further, theinterference imposed by the other cochannel MSs on the MSof interest may be increased due to increasing their respectivetransmit power for the sake of maintaining their link quality.However, in the absence of power control, the SIR of theMSs roaming far away from the RAs is seriously reduced inthe cell-edge area, particularly in the θworst direction, wherewe have 0 dB < SIRMS < 10 dB. As an improvement of userfairness, when applying power control in the θworst direction,the cell-edge MSs that suffer from a low SIR of SIRMS < 10 dBhave seen their average SIR increased to SIR′

MS ≈ 15 dB, asevidenced by Fig. 10.

Both the BER and the effective throughput of the MSssupported by the noncooperative-PDA-based FFR-DAS andthe SC-PDA-based MR-FFR-DAS along the θbest and θworst

directions in the presence of power control are characterizedin Fig. 11. More specifically, when using power control in theabsence of MRs, namely, when using the noncooperative PDA,the MSs roaming in the cell-edge area are capable of achievingan improved BER of about 10−3 in the θworst direction. Bycontrast, when invoking power control and selecting the MRsfrom the reliable area for the SC-PDA, the MSs roamingin the cell-edge area are capable of achieving BER < 10−4,even when they are in the θworst direction and quite close tothe cell-edge boundary, as shown in Fig. 11(a). Additionally,when considering the effective throughput, we observe fromFig. 11(b) that the SC-PDA-based MR-FFR-DAS significantlyoutperforms the PDA-based noncooperative FFR-DAS in mostof the cell-edge area, including the cell-edge boundary of boththe θbest and θworst directions. Meanwhile, in the SC-PDA-based MR-FFR-DAS, all the cell-edge MSs have achieved asimilarly high effective throughput, which implies having anappealing user fairness.

To more clearly demonstrate the benefits of jointly using thepower control and the MR-aided SC-PDA detector, the QoS dis-tribution across the entire cell-edge area is shown Fig. 12, wherewe define the minimum QoS required as SIR′

MS > 15 dB. Ourobservation area is defined as the 30◦ sector spanning the θbestand θworst directions, as visualized in Fig. 1, where a total ofNm = 6 MSs are simultaneously supported across the entirecell-edge area. We can see that, when the MSs roam close tothe RAs, all the six MSs are capable of achieving the QoStarget of SIR′

MS > 15 dB with the aid of power control, whichcorresponds to 63% of the entire cell-edge area, as shown inFig. 12. When the MSs are roaming close to the θworst direction,the number of MSs achieving the QoS target of SIR′

MS > 15 dBis reduced to 3, which corresponds to 24% of the cell-edge area.

Finally, in Fig. 13(a) and (b), we compared the SC-PDA-aided CoMP-CAS and the SC-PDA-aided MR-FFR-DAS interms of their achievable cell-edge BER and cell-edge ef-fective throughput, respectively. These results have clearlyshown that the SC-PDA-aided MR-FFR-DAS constitutes amore promising solution for improving the cell-edge perfor-mance of interference-limited multicell systems.

VI. CONCLUSION

In this paper, we have studied the achievable uplink cell-edgeperformance of four multicell system architectures, includingthe noncooperative CAS, the BS-cooperation-aided CoMP-CAS, the noncooperative FFR-DAS, and the MR-FFR-DASarchitectures. By benchmarking against three representativenoncooperative MUD schemes, we demonstrated that, typi-cally, both the CoMP-CAS and the MR-FFR-DAS relying onthe proposed SC-PDA receiver significantly outperform theirrespective noncooperative counterparts in terms of the BER ofthe cell-edge MSs. We revealed, however, that the performancegain achieved by the CoMP-CAS may be eroded when thetransmit power of the cell-edge MSs is low, and the MR-FFR-DAS is in general more effective in supporting cell-edge MSs


than the CoMP-CAS. We also demonstrated that the cell-edgeBER of the MR-FFR-DAS may be further reduced by usingjudicious MR selection, particularly for the MSs roaming in theworst-case direction or close to the cell-edge boundary. On theother hand, when considering the cell-edge effective throughputcalculated relying on the packet error rate, we showed thatthe SC-PDA-aided MR-FFR-DAS architecture does not alwaysoutperform its noncooperative counterpart, since the formerinvokes two time slots for completing a single MS transmission.Furthermore, we demonstrated that the effective throughput atthe cell edge and the fairness among the cell-edge MSs in theSC-PDA-aided MR-FFR-DAS may be significantly improvedby using multiuser power control. As a result, for low/moderateMS transmit power, the proposed SC-PDA-aided MR-FFR-DAS scheme is capable of achieving both a significantly betterBER and an improved effective throughput across the entirecell-edge area than the CoMP-CAS scheme.

REFERENCES

[1] J. G. Andrews et al., “What will 5G be?” IEEE J. Sel. Areas Commun.,vol. 32, no. 6, pp. 1065–1082, Jun. 2014.

[2] A. Ghosh, R. Ratasuk, B. Mondal, N. Mangalvedhe, and T. Thomas,“LTE-advanced: Next-generation wireless broadband technology,” IEEEWireless Commun., vol. 17, no. 3, pp. 10–22, Jun. 2010.

[3] 3GPP TSG RAN, Coordinated Multi-Point Operation for LTE PhysicalLayer Aspects (Rel. 11), 3GPP Std. TR 36.814 v.0.4.1 Std., Sep. 2013.[Online]. Available: http://www.3gpp.org/DynaReport/36819.htm

[4] M. Sawahashi, Y. Kishiyama, A. Morimoto, D. Nishikawa, andM. Tanno, “Coordinated multipoint transmission/reception techniques forLTE-advanced,” IEEE Wireless Commun., vol. 17, no. 3, pp. 26–34,Jun. 2010.

[5] D. Gesbert et al., “Multi-cell MIMO cooperative networks: A new look atinterference,” IEEE J. Sel. Areas Commun., vol. 28, no. 9, pp. 1380–1408,Dec. 2010.

[6] S. Yang, T. Lv, R. G. Maunder, and L. Hanzo, “Distributed probabilistic-data-association-based soft reception employing base station cooperationin MIMO-aided multiuser multicell systems,” IEEE Trans. Veh. Technol.,vol. 60, no. 7, pp. 3532–3538, Sep. 2011.

[7] P. Marsch and G. P. Fettweis, Coordinated Multi-Point in Mobile Com-munications. Cambridge, U.K.: Cambridge Univ. Press, 2011.

[8] R. Irmer et al., “Coordinated multipoint: Concepts, performance, andfield trial results,” IEEE Commun. Mag., vol. 49, no. 2, pp. 102–111,Feb. 2011.

[9] D. Lee et al., “Coordinated multipoint transmission and reception inLTE-advanced: Deployment scenarios and operational challenges,” IEEECommun. Mag., vol. 50, no. 2, pp. 148–155, Feb. 2012.

[10] S. A. Ramprashad et al., “Cooperative cellular networks using multi-userMIMO: Trade-offs, overheads, and interference control across architec-tures,” IEEE Commun. Mag., vol. 49, no. 5, pp. 70–77, May 2011.

[11] A. Adhikary, J. Nam, J.-Y. Ahn, and G. Caire, “Joint spatial divisionand multiplexing-the large-scale array regime,” IEEE Trans. Inf. Theory,vol. 59, no. 10, pp. 6441–6463, Oct. 2013.

[12] H. Ekstrom et al., “Technical solutions for the 3G long-term evolution,”IEEE Commun. Mag., vol. 44, no. 3, pp. 38–45, Mar. 2006.

[13] N. Saquib, E. Hossain, and D. I. Kim, “Fractional frequency reusefor interference management in LTE-advanced hetnets,” IEEE WirelessCommun., vol. 20, no. 2, pp. 113–122, Apr. 2013.

[14] T. D. Novlan, R. K. Ganti, A. Ghosh, and J. G. Andrews, “Analyti-cal evaluation of fractional frequency reuse for OFDMA cellular net-works,” IEEE Trans. Wireless Commun., vol. 10, no. 12, pp. 4294–4305,Dec. 2011.

[15] X. Li, T. Jiang, S. Cui, J. An, and Q. Zhang, “Cooperative communicationsbased on rateless network coding in distributed MIMO systems,” IEEEWireless Commun., vol. 17, no. 3, pp. 60–67, Jun. 2010.

[16] O. Simeone, O. Somekh, Y. Bar-Ness, and U. Spagnolini, “Up-link throughput of TDMA cellular systems with multicell processingand amplify-and-forward cooperation between mobiles,” IEEE Trans.Wireless Commun., vol. 6, no. 8, pp. 942–2951, Aug. 2007.

[17] H. Zhang, N. B. Mehta, A. F. Molisch, J. Zhang, and H. Dai, “Asynchro-nous interference mitigation in cooperative base station systems,” IEEETrans. Wireless Commun., vol. 7, no. 1, pp. 155–165, Jan. 2008.

[18] G. Caire, S. A. Ramprashad, and H. C. Papadopoulos, “Rethinking net-work MIMO: Cost of CSIT, performance analysis, and architecture com-parisons,” in Proc. ITA Workshop, San Diego, CA, USA, Jan. 2010,pp. 1–10.

[19] K. Hosseini, W. Yu, and R. S. Adve, “Large-scale MIMO versus networkMIMO for multicell interference mitigation,” IEEE J. Sel. Topics SignalProcess., vol. 8, no. 5, pp. 930–941, Oct. 2014.

[20] H. Lei, L. Zhang, X. Zhang, and D. Yang, “A novel multi-cell OFDMAsystem structure using fractional frequency reuse,” in Proc. IEEE 18th Int.Symp. PIMRC, Athens, Greece, Sep. 2007, pp. 1–5.

[21] Y. Xiang, J. Luo, and C. Hartmann, “Inter-cell interference mitiga-tion through flexible resource reuse in OFDMA based communica-tion networks,” in Proc. Eur. Wireless Conf., Paris, France, Apr. 2007,pp. 1–7.

[22] A. L. Stolyar and H. Viswanathan, “Self-organizing dynamic frac-tional frequency reuse in OFDMA systems,” in Proc. 27th IEEE Conf.INFOCOM, Phoenix, AZ, USA, Apr. 2008, pp. 1364–1372.

[23] X. Xu, R. Zhang, S. Ghafoor, and L. Hanzo, “Imperfect digital fibreoptic link based cooperative distributed antennas with fractional frequencyreuse in multicell multiuser networks,” IEEE Trans. Veh. Technol., vol. 60,no. 9, pp. 4439–4449, Nov. 2011.

[24] D. Wake, M. Webster, G. Wimpenny, K. Beacham, and L. Crawford,“Radio over fiber for mobile communications,” in Proc. IEEE Int. TopicalMWP, Ogunquit, ME, USA, Oct. 2004, pp. 157–160.

[25] W. Choi and J. G. Andrews, “Downlink performance and capacity ofdistributed antenna systems in a multicell environment,” IEEE Trans.Wireless Commun., vol. 6, no. 1, pp. 69–73, Jan. 2007.

[26] A. K. Sadek, Z. Han, and K. J. R. Liu, “Distributed relay-assignmentprotocols for coverage expansion in cooperative wireless networks,” IEEETrans. Mobile Comput., vol. 9, no. 4, pp. 505–515, Apr. 2010.

[27] T. Mayer, H. Jenkac, and J. Hagenauer, “Turbo base-station coopera-tion for intercell interference cancellation,” in Proc. IEEE ICC, Istanbul,Turkey, Jun. 2006, pp. 4977–4982.

[28] S. Khattak, W. Rave, and G. Fettweis, “Distributed iterative multiuser de-tection through base station cooperation,” EURASIP J. Wireless Commun.Netw., vol. 2008, p. 17, Feb. 2008.

[29] E. Aktas, J. Evans, and S. Hanly, “Distributed decoding in a cellularmultiple-access channel,” IEEE Trans. Wireless Commun., vol. 7, no. 1,pp. 241–250, Jan. 2008.

[30] D. Radji and H. Leib, “Interference cancellation based detection forV-BLAST with diversity maximizing channel partition,” IEEE J. Sel.Topics Signal Process., vol. 3, no. 6, pp. 1000–1015, Dec. 2009.

[31] J. Lee, J.-W. Choi, H.-L. Lou, and J. Park, “Soft MIMO ML demodulationbased on bitwise constellation partitioning,” IEEE Commun. Lett., vol. 13,no. 10, pp. 736–738, Oct. 2009.

[32] S. Yang, T. Lv, R. Maunder, and L. Hanzo, “From nominal to true aposteriori probabilities: An exact Bayesian theorem based probabilisticdata association approach for iterative MIMO detection and decoding,”IEEE Trans. Commun., vol. 61, no. 7, pp. 2782–2793, Jul. 2013.

[33] E. Katranaras, M. Imran, and C. Tzaras, “Uplink capacity of a variabledensity cellular system with multicell processing,” IEEE Trans. Commun.,vol. 57, no. 7, pp. 2098–2108, Jul. 2009.

[34] D. Wake, A. Nkansah, and N. J. Gomes, “Radio over fiber link designfor next generation wireless systems,” J. Lightw. Technol., vol. 28, no. 16,pp. 2456–2464, Aug. 2010.

[35] K. Y. Lau, “RF transport over optical fiber in urban wireless infrastruc-tures,” IEEE/OSA J. Opt. Commun. Netw., vol. 4, no. 4, pp. 326–335,Apr. 2012.

[36] G. P. Agrawal, Nonlinear Fiber Optics, 4th ed. New York, NY, USA:Academic, 2006.

[37] T. S. Rappaport and L. B. Milstein, “Effects of radio propagation pathloss on DS-CDMA cellular frequency reuse efficiency for the reversechannel,” IEEE Trans. Veh. Technol., vol. 41, no. 3, pp. 231–242,Aug. 1992.

[38] V. Sreng, H. Yanikomeroglu, and D. D. Falconer, “Relayer selectionstrategies in cellular networks with peer-to-peer relaying,” in Proc. IEEE58th VTC -Fall, Orlando, FL, USA, Oct. 2003, pp. 1949–1953.

[39] R. U. Nabar, H. Bolcskei, and F. W. Kneubuhler, “Fading relay channels:Performance limits and space-time signal design,” IEEE J. Sel. AreasCommun., vol. 22, no. 6, pp. 1099–1109, Aug. 2004.

[40] P. Kyritsi, D. C. Cox, R. A. Valenzuela, and P. W. Wolniansky, “Cor-relation analysis based on MIMO channel measurements in an indoorenvironment,” IEEE J. Sel. Areas Commun., vol. 21, no. 5, pp. 713–720,Jun. 2003.


[41] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.Cambridge, U.K.: Cambridge Univ. Press, 2005.

[42] F. D. Neeser and J. L. Massey, “Proper complex random processes withapplications to information theory,” IEEE Trans. Inf. Theory, vol. 39,no. 4, pp. 1293–1302, Jul. 1993.

[43] T. Adali, P. Schreier, and L. Scharf, “Complex-valued signal processing:The proper way to deal with impropriety,” IEEE Trans. Signal Process.,vol. 59, no. 11, pp. 5101–5125, Nov. 2011.

[44] D. Mandic and S. L. Goh, Complex Valued Nonlinear Adaptive Fil-ters: Noncircularity, Widely Linear and Neural Models, ser. Adaptive andLearning Systems for Signal Processing, Communications and ControlSeries. Hoboken, NJ, USA: Wiley, 2009.

[45] L. Hanzo, P. Cherriman, and J. Streit, Wireless Video Communications:Second to Third Generation and Beyond. Hoboken, NJ, USA: Wiley-IEEE Press, 2001.

[46] 3GPP TSG-RAN, Technical specification group radio access network;Further Advancements for E-UTRA: Physical Layer Aspects, 3GPP Std.TR 36.814 v.0.4.1 Std., 2009.

Shaoshi Yang (S’09–M’13) received the B.Eng. de-gree in information engineering from Beijing Uni-versity of Posts and Telecommunications (BUPT),Beijing, China, in 2006; the Ph.D. degree in electron-ics and electrical engineering from the University ofSouthampton, Southampton, U.K., in 2013; and thePh.D. degree in signal and information processingfrom BUPT in 2014.

From 2008 to 2009, he was an Intern ResearchFellow with Intel Labs China, Beijing, where he fo-cused on channel quality indicator channel design for

mobile WiMAX (802.16m). Since 2013, he has been a Postdoctoral ResearchFellow with the University of Southampton. He has authored or coauthoredmore than 30 research papers for IEEE journals and conferences. His researchinterests include multiple-input–multiple-output signal processing, green radio,heterogeneous networks, cross-layer interference management, and convexoptimization and its applications (https://sites.google.com/site/shaoshiyang/).

Dr. Yang has served as a Technical Program Committee Member for anumber of IEEE conferences and journals, including IEEE ICC, PIMRC,ICCVE, and HPCC, as well as the IEEE JOURNAL ON SELECTED AREAS INCOMMUNICATIONS. He is also a Junior Member of the Isaac Newton Institutefor Mathematical Sciences, University of Cambridge, Cambridge, U.K. Hehas received a number of academic and research awards, including the PMC-Sierra Telecommunications Technology Scholarship at BUPT, the Electronicsand Computer Science Scholarship of the University of Southampton, and theBest Ph.D. Thesis Award from BUPT.

Xinyi Xu received the B.Eng. degree in electronic in-formation from Wuhan University, Wuhan, China, in2004; the Ph.D. degree in wireless communicationsfrom the University of Southampton, Southamp-ton, U.K., in 2013 (which was sponsored by theU.K./China Scholarships for Excellence Programmefrom 2008); and the Ph.D. degree in communica-tions and information system from Wuhan Universityin 2014.

She is currently with the China Academy of Elec-tronics and Information Technology, Beijing, China.

Her research interests include future networks and integrated space aroundnetworks.

Dimitrios Alanis (S’13) received the M.Eng. degreein electrical and computer engineering from AristotleUniversity of Thessaloniki, Thessaloniki, Greece,in 2011 and the M.Sc. degree in wireless com-munications from the University of Southampton,Southampton, U.K., in 2012. He is currently work-ing toward the Ph.D. degree with the SouthamptonWireless Group, School of Electronics and ComputerScience, University of Southampton.

His research interests include quantum compu-tation and quantum information theory, quantum

search algorithms, cooperative communications, resource allocation for self-organizing networks, bioinspired optimization algorithms, and classical andquantum game theory.

Soon Xin Ng (S’99–M’03–SM’08) received theB.Eng. degree (with first-class honors) in elec-tronic engineering and the Ph.D. degree in telecom-munications from the University of Southampton,Southampton, U.K., in 1999 and 2002, respectively.

From 2003 to 2006, he was a Postdoctoral Re-search Fellow working on collaborative Europeanresearch projects known as SCOUT, NEWCOM, andPHOENIX. Since August 2006, he has been a Mem-ber of Academic Staff with the School of Electronicsand Computer Science, University of Southampton,

where he is currently an Associate Professor in telecommunications. Heis involved in the OPTIMIX and CONCERTO European projects and theIU-ATC and UC4G projects. His research interests include adaptive coded mod-ulation, coded modulation, channel coding, space–time coding, joint source andchannel coding, iterative detection, orthogonal frequency division multiplex-ing, multiple-input multiple-output, cooperative communications, distributedcoding, quantum error correction codes, and joint wireless and optical fibercommunications. He has authored or coauthored over 180 papers and two JohnWiley/IEEE Press books in these fields.

Dr. Ng is a Fellow of the Higher Education Academy in the U.K. He is alsoa Chartered Engineer.

Lajos Hanzo (M’91–SM’92–F’04) received theMaster’s degree in electronics, the Ph.D. degree, andthe Doctor Honoris Causa degree from the TechnicalUniversity of Budapest, Budapest, Hungary, in 1976,1983, and 2009, respectively, and the D.Sc. degreefrom the University of Southampton, Southampton,U.K., in 2004.

During his career in telecommunications, hehas held various research and academic posts inHungary, Germany, and the U.K. Since 1986, he hasbeen with the School of Electronics and Computer

Science, University of Southampton, Southampton, U.K., where he holds theChair in telecommunications. He was a Chaired Professor with TsinghuaUniversity, Beijing, China. He is the coauthor of 20 John Wiley/IEEE Pressbooks on mobile radio communications, totalling in excess of 10 000 pages, andhas published more than 1400 research entries on IEEE Xplore. He is currentlydirecting an academic research team, working on a range of research projectsin the field of wireless multimedia communications sponsored by industry,the Engineering and Physical Sciences Research Council (EPSRC) U.K., theEuropean IST Program, and the Mobile Virtual Centre of Excellence, U.K. Heis an enthusiastic supporter of industrial and academic liaison, and he offers awide range of industrial courses.

Dr. Hanzo has acted as a Technical Program Committee Chair for IEEE con-ferences, presented keynote lectures, and has received a number of distinctions.He is the Governor of the IEEE Vehicular Technology Society and the PastEditor-in-Chief of the IEEE Press.

is the low-complexity mobile-relay-aided ffr-das capable ...sxn/paper/201604tvt.pdf · cooperation,...

Documents